Finding the queries an audience uses, sizing the demand and mapping them to pages. In AI search the unit shifts from a keyword to a prompt and the queries it fans out into.
Keyword Research is the practice of finding the queries an audience types, sizing the demand behind them, and mapping them to pages. The output is a list of queries with volume, competition and intent, grouped so that one page targets one cluster rather than one string.
Search Console supplies the queries a site already appears for with impressions, clicks and average position, and it is the only source tied to the site's own performance. Keyword tools supply volume estimates and competitor coverage. Autocomplete, related searches and question feeds supply phrasing. Clustering the resulting list on embeddings groups paraphrases that string matching keeps apart.
A generative engine does not answer the typed query alone. It expands it into several sub-queries through query fan-out, retrieves for each, and composes one answer, so the unit of research moves from a keyword to a prompt and the set of queries it generates. Volume for those generated queries is not reported anywhere, which is the gap LLM search volume prediction addresses, and the reason implicit queries are worth mapping.
